OPINION
MURGUIA, Circuit Judge:
Petitioner Manuel Tarango, Jr. appeals the district court’s denial of his petition for a writ of habeas . corpus. He claims violation of his due process right to a fair and impartial jury, where a police vehicle followed Juror No. 2, a known holdout against a guilty verdict, for approximately seven miles, on the second day of deliberations, in a highly publicized trial involving multiple police victims. Tarango argues that the Nevada Supreme Court’s decision upholding his convictions “was contrary to, or involved an unreasonable application of, clearly established federal law,” see 28 U.S.C. § 2254(d)(1), because the court failed to consider whether the contact be[940]*940tween the juror and the police vehicle prejudiced the jury’s verdict.
We hold that the Nevada Supreme Court’s decision was contrary to Mattox v. United States, 146 U.S. 140, 13 S.Ct. 50, 36 L.Ed. 917 (1892), because the court improperly- limited its inquiry to whether the external contact amounted to a “communication” and did not investigate the prejudicial effect of the police tail. We therefore review de novo the question whether the extrinsic contact could have influenced the verdict and prejudiced Tarango. Because the trial court prevented Tarango from offering certain evidence to demonstrate prejudice, we remand for an evidentiary hearing and further fact finding.
BACKGROUND
On December 5, 1999, a rock band of off-duty Las Vegas police officers,. Metro Mike’s Pigs in a Blanket, was performing at a local bar called Mr. D’s. The bar was filled with off-duty police officers. A group of masked men entered the bar announcing a robbery, and a shoot-out ensued. Several patrons were shot, one robber was shot and killed, and one police officer, Officer Dennis Devitte, was shot several times. The surviving robbers escaped the scene and, six years later, Tarango was brought to trial on seven felony counts. The 2005 trial received considerable local media attention, and numerous Las Vegas Metro police officers attended as both witnesses and spectators.
After the jury began its deliberations, on November 1, 2005, the foreperson sent a note to the trial judge indicating that the jury had “reached a stalemate” because of a “problem juror” who had “made it very clear he does not want to be part of [the] process [and] is refusing to discuss or interact with the other jurors.” The “problem juror” separately wrote to the judge indicating that he had “doubt of which [he] feel[s] is beyond the limit of reasonable doubt,” and that deliberations were “not curing [his] doubt.” In his note, the “problem” juror identified himself as Juror No. 2.
Over Tarango’s objection,1 the judge advised the jury to continue deliberating. The next day, November 2nd, the jury returned a verdict finding Tarango guilty of all seven felony counts as charged: burglary with the use of a deadly weapon, attempted robbery with the use of a deadly weapon, conspiracy to commit robbery with the use of a deadly weapon, three counts of battery with the use of a deadly weapon, and attempted murder with the use of a deadly weapon- — all in violation of Nevada state law.
On November 3rd, the Las Vegas Review-Journal reported the guilty verdict in an article titled Man Convicted in 1999 Case. The article referenced “a juror who spoke to the Review-Journal.” Discussing the jury’s deliberation process, the interviewed juror mentioned the hold-out juror: “the case was close to a hung jury because one juror seemed.unwilling to convict following nearly two days of deliberations.”
On November 4th, prompted by the previous day’s newspaper article, Juror No. 2 wrote a letter to the court referencing the article:
I am the one Juror mentioned in the article.... I am also the Juror that wrote you the note during deliberations. It read: “I have doubt beyond the limit of what I consider reasonable doubt.” I also stated, “I did not believe further deliberations would cure that doubt.” [941]*941Further deliberations in fact, did not cure my doubt.
However, when returning to re-deliberate Wednesday November 2nd from the Henderson area, a Metro squad car followed me northbound on 1-95 and into the downtown area.
I found that action unnerving.
I realize the State has much time and money invested in this case. There were [sic] no alternate Juror. I concluded Metro somehow knew who I was and knew of my unwillingness to convict. I have never been in trouble with the law. Therefore, I relinquished my vote under duress. I only ask, within the law, please show [Tarango] leniency.
One week later, on November 11th, Juror No. 2 emailed Tarango’s trial attorney, Marc Saggese, and attached a copy of his “Letter to the Judge.” The juror told Sag-gese that he felt “compelled to notify” Saggese of the letter. Saggese promptly filed a motion to dismiss all charges with prejudice or, alternatively, to grant a new trial on the ground of juror misconduct, arguing that Juror No. 2’s communication indicated that the deliberation process had been tampered with in violation of Taran-go’s right to due process. • Under- Nevada law, juror misconduct refers to two categories of conduct: (1) intrinsic misconduct, that is, “conduct by jurors contrary to them instructions or oaths;” and (2) extrinsic misconduct, or “attempts' by third parties to influence the jury process.”2 Meyer v. State, 119 Nev. 554, 80 P.3d 447, 453 (2003). Tarango alleged both forms of misconduct, arguing that..(1) Juror No. 2 changed his vote, under pressure, rather than based, on- admissible evidence of Tar-ango’s guilt, because of (2) an improper third party influence.
In support of the motion, Saggese submitted a declaration indicating that, after the trial court-read the juror notes into the record and while deliberations were ongoing, Saggese overheard Deputy District Attorney Marc DiGiacomo report to Detective James Vacarro over the phone that one juror, Juror No. 2,'was holding out. Saggese thus indirectly corroborated Juror No. 2’s stated belief that he was being targeted as a hold-out juror by introducing evidence that members of the Las Vegas police department both knew that Juror No. 2 favored acquittal and had knowledge of Juror No. 2’s identity.3
The trial court held a full hearing on Tarango’s motion the following month. Juror No. 2, Defense Attorney Saggese, Detective Vacarro, and Deputy D.A. DiGiaco-mo were all called to testify regarding their knowledge of the alleged events and communications in question. At the hearing, the court limited the questioning of [942]*942Juror No. 2 pursuant to a provision of the Nevada Code of Evidence, Nev. Rev. Stat. § 50.065, which prohibits the admission for any purpose of testimony, affidavits, or evidence of any statement by a juror indicating an effect on the jury’s deliberative process. The court also relied on the Nevada Supreme Court case of Meyer v. State, which provides that “[u]pon an inquiry into the validity of a verdict ..., a juror may not testify as to any matter or statement occurring during the course of the jury’s deliberations, or to the effect of anything upon that or any other juror’s mind.”4 80 P.3d at 454 (quoting Fed. R. Evid. 606(b)). The trial court ultimately conducted all questioning of Juror No. 2 itself. Juror No. 2 testified as follows:
[Right after getting on the freeway,] I was in the center lane [of US-95]. I noticed a Metro squad car behind me; fairly close behind me.... He was close enough I couldn’t see his front-wheels or bumper. And, I looked down and I was not exceeding the speed limit.
[[Image here]]
I signaled and got over to the far right lane anticipating being pulled over and he stayed tight behind me.
[[Image here]]
I maintained under the speed limit anticipating being pulled over. A couple minutes and he never lit up, he never indicated that he was ... going to pull me over. So I just maintained right lane position under the speed limit. This continued on.
[[Image here]]
[At Eastern Avenue] there was a lot of traffic entering the freeway.... [T]here was so many cars trying to merge into the freeway that the Metropolitan squad car actually pulled up closer to prevent anyone from pulling in between our vehicles.
[[Image here]]
And as soon as the ... exit to Las Vegas Boulevard came, I even slowed down under 50, and that’s a long exit there. It’s, um, a quarter mile, half a mile, and even at that, he maintained position.
And he’s not pulling me over. He’s not ... giving me a citation for nothing. He followed me down the hill, and at the stoplight for Lás Vegas Boulevard.... He followed me, still tight. And there’s several stop lights, something, Stewart, and then Carson is where the juror parking garage is. And we did get a red light there. He was still behind me. I took a right to enter the ... jurors parking lot. That’s when he relieved me from the escort or whatever he was doing. That’s when he left me alone.
When questioned, Juror No. 2 indicated that he could not tell whether the driver of the vehicle was male or female, and he could not report the squad car number. However, Juror No. 2 averred that the car behind him was “a Metropolitan black and white vehicle.” When questioned a second time, Juror No. 2 reiterated that the car remained “consistently” tight behind him for the duration of his commute to the [943]*943courthouse — “[c]lose enough that [he] couldn’t see the officer’s bumper.”
At the end of the hearing, the court orally denied Tarango’s motion to dismiss or to grant a new trial. The trial court did not discredit Juror No. 2’s testimony, and made one factual finding that Juror No. 2 “was followed closely, tightly, however you want to state it from Tropicana on US-95 to Las Vegas Boulevard and Carson.”5 The court went on to reach the following legal conclusion:
I don’t think there’s any evidence of juror misconduct. There were no attempts to influence the jury. There’s no outside influence on this particular juror. There’s no communication or contact. The alleged conduct is ambiguous, it’s vague and nonspecific in content. I’m required to consider this extrinsic influence in light of the trial as a whole, and consider the weight of the evidence against Mr. Tarango and with that, and based on the [Meyer] decision, and the reasonable person test that I’m required to apply. I don’t think that Mr. Tarango has met his burden. Therefore, the motion is denied.6
Weeks later, at a televised proceeding on February 8, 2006, the trial court denied Tarango’s motion to reconsider on the basis of jury misconduct and entered judgment against him. The trial court sentenced Tarango to a 22-58 year term of imprisonment. Tarango promptly appealed the denial.
In September 2007, the Nevada Supreme Court affirmed the state trial court’s denial of Tarango’s motion for a new trial. Tarango v. State, 123 Nev. 864, 210 P.3d 773 (2007). The Nevada Supreme Court stated the relevant test as follows: “For a defendant to prevail on a motion for a new trial based on misconduct, the defendant must present admissible evidence sufficieiit to establish (1) the occurrence of misconduct, and (2) a showing that the misconduct was prejudicial.”7 Id., slip op. at 2 (citing Meyer, 80 P.3d at 455).
The Nevada Supreme Court first concluded that Juror No. 2’s letters to the trial court were properly deemed inadmissible to prove that Juror No. 2 had voted guilty in violation of the jury instructions or contrary to his oath as a juror, reasoning that “for misconduct to be proved it ‘must be based on objective facts and not the state of mind or deliberative process of the jury.’ ” Id. slip op. at 6 (quoting Meyer, 80 P.3d at 454). Absent Juror No. 2’s letters, the Nevada Supreme Court concluded that the “testimony of [Defense Attorney Saggese, Detective Vacarro, and [944]*944Deputy D.A. DiGiacomo] was insufficient to show by objective facts that [Juror No. 2] committed misconduct.” Id. The Nevada Supreme Court held that Tarango had thus failed to show by admissible evidence that Juror No. 2 had committed misconduct. Id.
The Nevada Supreme Court further held that there was no evidence of an improper external influence on Juror No. 2. Although the Nevada Supreme Court assumed “arguendo that [Juror, No. 2] was followed by a marked police car,” and, observed that “any unauthorized communication between law enforcement and a juror about a matter pending before a jury may be ‘presumptively prejudicial,’ ” the court concluded that “[Juror No. 2] failed to show by objective facts that there was an improper external communication between him and the police.” Id. slip op. at 6-7. The Nevada Supreme Court explained that “it is not clear whether being followed by a marked car qualifies as a communication at all. It is even more dubious as to whether such a ‘communication’ was about a matter pending before the jury.” Id. slip op. at 7. 'In other words, having found that no “communication” had occurred, the Nevada Supreme Court determined that the alleged influence of the non-communicative contact was “too speculative” to sustain Tarango’s motion for a new trial and did not reach the second prong of the misconduct inquiry — whether the contact was prejudicial. Id.
Following state habeas proceedings, Tarango timely filed his federal habeas petition on .March 15, 2010. The federal district court for the District of Nevada was “tempted to say that the fact Juror 2 rendered his verdict based not upon the law and evidence, but because of his perception of a threat, is dispositive.” However, without citing authority, the district court concluded that “Supreme Court case law is clear that objective proof of external contact is required.” It further concluded that-the state court did not err in concluding that no external contact had occurred, although the. court found that determination “debatable.”'The district court therefore dismissed Tarango’s petition in September 2013, upholding as reasonable the state court’s determination that Tarango had failed to show any improper external contact.
On October 16, 2013, the district court granted Tarango a Certificate of Appeala-bility" as to Ground One of his amended petition, and Tarango filed a Notice of Appeal the same day. Ground One reads as follows:
Tarango was convicted because one of the jurors believed that the State was trying to intimidate him, and not because he believed Tarango was guilty. As such, Tarango is incarcerated in violation of his right to a Fair Trial, an Impartial Jury, and Due Process under the 6th and 14th Amendments of the United States Constitution.
Tarango raises only the certified issue in his appeal before us.
DISCUSSION
I.
We review de novo a district court’s denial of a habeas corpus petition. Hurles v. Ryan, 752 F.3d 768, 777 (9th Cir. 2014). But where, as here, a state court has adjudicated a claim on the merits, the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA) compels us to accord significant deference to the underlying state court decision. See 28 U.S.C. § 2254(d)(1)—(2). This court may grant relief only when the state court’s adjudication of that claim either (1) “resulted in a decision that was contrary to, or involved an unreasonable application of, clearly es[945]*945tablished federal law, as determined by the Supreme Court of the United States,” or (2) was “based on an unreasonable determination of the facts in light , of the evidence presented at the State Court proceeding.” Id.
Where a.state court fails to apply the clearly established federal law, applying an incorrect standard in reaching its decision, “the state court’s adjudication [is] contrary to clearly established law.” Lafler v. Cooper, — U.S. -, 132 S.Ct. 1376, 1390, 182 L.Ed.2d 398 (2012) (holding that state court adjudication was contrary to clearly established federal law because it failed to apply Strickland to an ineffeetive-assistance-of-counsel claim). And in that circumstance, federal habeas courts “can determine the principles necessary to grant relief.” Id. (citing Panetti v. Quarterman, 551 U.S. 930, 948, 127 S.Ct. 2842, 168 L.Ed.2d 662 (2007)). In other words, a state court’s failure to apply the' proper standard under clearly established federal law “allows federal-court review ... without deference to the state court’s decision” and “unencumbered by the deference AEDPA normally requires.” Panetti, 551 U.S. at 948, 127 S.Ct. 2842; see also Castellanos v. Small, 766 F.3d 1137, 1146 (9th Cir. 2014) (“If the state court applies a legal standard .that contradicts clearly established federal law, we review de novo the applicant’s claims, applying the correct legal standard to determine whether the applicant is entitled to relief.” (citing Cooperwood v. Cambra, 245 F.3d 1042, 1047 (9th Cir. 2001))).
In conducting this review, we look to the “last reasoned decision” by a state court addressing the issue at hand. Miles v. Ryan, 713 F.3d 477, 486 (9th Cir. 2012) (citing Robinson v. Ignacio, 360 F.3d 1044, 1055 (9th Cir. 2004)). In this case, we look to the Nevada Supreme Court’s September 2007 decision affirming the state trial court’s judgment on direct appeal.
II.
The Nevada Supreme Court, after assuming that Juror No, 2 was followed by a police car, decided that such contact did not implicate Tarango’s right to due process because it did not amount to a “communication,” much less a communication “about a matter pending before the jury.” The court declined to consider whether the police tail could have prejudiced the verdict, We hold that the Nevada Supreme Court violated clearly established Supreme Court case law, first by limiting its inquiry to whether the contact amounted to a “communication ... about a matter pending before the jury” and, second, by failing to examine the potential impact of the non-communicative contact on Juror No. 2’s verdict.
A.
In criminal trials, well-entrenched Supreme Court authority “absolutely” forbids “external causes tending to disturb the [jury’s] exercise of deliberate and unbiased judgment ... at least until their harmlessness is made to appear.” Mattox v. United States, 146 U.S. 140, 149-50, 13 S.Ct. 50, 36 L.Ed. 917 (1892). We have held that Mattox established a bright-line rule: any external contact with a juror is subject to a presumption that the contact prejudiced the jury’s verdict, but the government may overcome that presumption by showing that the contact was harmless. Caliendo v. Warden of Cal. Men’s Colony, 365 F.3d 691, 696 (9th Cir. 2004) (citing United States v. Armstrong, 654 F.2d 1328, 1331-33 (9th Cir. 1981)).
Clearly established federal law provides that any unauthorized “private communication, contact, or tampering di[946]*946rectly or indirectly, with a juror during a trial about the matter pending before the jury is, for obvious reasons, deemed presumptively prejudicial.” Remmer v. United States, 347 U.S. 227, 229, 74 S.Ct. 450, 98 L.Ed. 654 (1954) (emphasis added). However, clearly established federal law also compels a criminal trial court to consider the prejudicial effect of any external contact that has a “tendency” to influence the verdict, irrespective of whether it is about the matter pending before the jury. Mattox, 146 U.S. at 150-51, 13 S.Ct. 50. Moreover, an external contact need not amount to a “communication” to trigger some judicial inquiry into possible prejudice. See Smith v. Phillips, 455 U.S. 209, 212-15, 102 S.Ct. 940, 71 L.Ed.2d 78 (1982) (requiring judicial inquiry into possible prejudice arising from a juror’s job application in the office of the prosecutor trying the case); Mattox, 146 U.S. at 150, 13 S.Ct. 50 (recognizing the prejudicial potential of “the reading of newspapers”).
B.
The Supreme Court has not established a bright-line test for determining what constitutes a possibly prejudicial “external” influence on a jury. The Court has devoted more recent attention to clarifying what “falls on the ‘internal’ side of the line.” Warger v. Shauers, — U.S. -, 135 S.Ct. 521, 529, 190 L.Ed.2d 422 (2014) (holding that a juror’s dishonesty during voir dire is internal to the deliberative process and not admissible to impeach a verdict); see also Tanner v. United States, 483 U.S. 107, 118-25, 107 S.Ct. 2739, 97 L.Ed.2d 90 (1987) (holding that jurors’ consumption of drugs and alcohol during trial is internal to the deliberative process and not admissible to impeach a verdict). It is clearly established that a juror’s physical or mental incapacity, substance abuse, and dishonesty during voir dire all amount to internal — not external — influences on a jury’s verdict. Tanner, 483 U.S. at 118-25, 107 S.Ct. 2739; Warger, 135 S.Ct. at 529. On the other end of the spectrum, the Court long ago explained that an “extraneous influence” would include “something which did not essentially inhere in the verdict, — an overt act, open to the knowledge of all the jury, and not alone within the personal consciousness of one.” Mattox, 146 U.S. at 149, 13 S.Ct. 50 (quoting Perry v. Bailey, 12 Kan. 539, 545 (1874)).
In more recent decisions interpreting the Mattox rule, the Court has clarified that an external contact need not be intentional, Gold v. United States, 352 U.S. 985, 77 S.Ct. 378, 1 L.Ed.2d 360 (1957) (granting a new trial where the FBI approached jurors about a different but related case, even though “the intrusion was unintentional”), nor verbal, Smith, 455 U.S. at 212-15, 221, 102 S.Ct. 940; see also Mattox, 146 U.S. at 150, 13 S.Ct. 50 (noting that the presence of an officer in the jury room during the deliberations would be “fatal to the verdict”). Rather, an impermissible external influence can arise where, for example, a juror is shown to have a relationship with the office of the prosecutor trying the case. Smith, 455 U.S. at 212, 102 S.Ct. 940. In addition, an external contact need only have influenced one juror, because a defendant is “entitled to be tried by 12 ... impartial and unprejudiced jurors.” Parker v. Gladden, 385 U.S. 363, 366, 87 S.Ct. 468, 17 L.Ed.2d 420 (1966).
The Supreme Court has identified an “extraneous influence” requiring judicial inquiry into prejudice in cases where the jury heard and read information about the defendant’s propensity for murder, which was not admitted into evidence, Mattox, 146 U.S. at 150-51, 13 S.Ct. 50; where members of a jury overheard the bailiff make disparaging' comments about the de[947]*947fendant, Parker, 385 U.S. at 363-65, 87 S.Ct. 468; where a juror was contacted by an FBI agent after being offered a bribe to acquit the defendant, Remmer, 347 U.S. at 228-30, 74 S.Ct. 450; and where a juror had submitted an application for employment at the office of the prosecutor trying the case, Smith, 455 U.S. at 212, 216-17, 102 S.Ct. 940.
C.
Mattox requires a trial court to examine possible prejudice when it is confronted with evidence of an external contact that has a “tendency” to be “injurious to the defendant.” Mattox, 146 U.S. at 150, 13 S.Ct. 50. Thus, an external contact with a juror need only raise a credible risk of influencing the verdict to be deemed possibly prejudicial. Mattox and its progeny further establish that undue contact with a juror by a government officer almost categorically risks influencing the verdict. Indeed, Mattox observed that the mere presence of a court officer or bailiff during the jury’s deliberations would “absolutely vitiate the verdict ... without regard to whether any improper influences were actually exerted over the jury or not.” Mattox, 146 U.S. at 150, 13 S.Ct. 50 (emphasis added); see also Smith, 455 U.S. at 221, 102 S.Ct. 940 (holding that a juror’s pending job application with the prosecutor’s office required a post-trial hearing on juror bias); Parker, 385 U.S. at 365, 87 S.Ct. 468 (“[T]he official character of the bailiff — as an officer of the court as well as the State — beyond question carries great weight with a jury....”); Remmer, 347 U.S. at 229, 74 S.Ct. 450 (“The sending of an FBI agent in the midst of a trial to investigate a juror as to his conduct is bound to impress the juror and is very apt to do so unduly.”).
To be sure, “it is virtually impossible to shield jurors from every contact or influence that might theoretically affect their vote.” Smith, 455 U.S. at 217, 102 S.Ct. 940. “[D]ue process does not require a new trial every time a juror has been placed in a compromising situation.” Id. Threadbare or speculative allegations, or allegations involving “prosaic kinds of jury misconduct” do not trigger a presumption of prejudice. United States v. Dutkel, 192 F.3d 893, 894-85 (9th Cir. 1999); see also Xiong v. Felker, 681 F.3d 1067, 1077 (9th Cir. 2012) (noting that the Supreme Court jurisprudence regarding juror misconduct “all involved ... significant, and in some cases deliberate interference with the deliberation process”).
Mindful of this reality, and given the need to preserve the finality of a jury’s verdict, courts universally prohibit jurors from impeaching their own verdicts through evidence of their internal deliberative process. See, e.g., Tanner, 483 U.S. at 117-20, 107 S.Ct. 2739. However, regardless of the forms of evidence admissible to demonstrate that a contact occurred, see United States v. Rutherford, 371 F.3d 634, 644-45 (9th Cir. 2004), the Supreme Court has unequivocally and repeatedly held that due process requires a trial judge to endeavor to “determine the effect” of occurrences tending to prejudice the jury when they happen.8 Smith, 455 U.S. at 217, 102 S.Ct. 940; see also Parker, 385 U.S. at 365, 87 S.Ct. 468; Remmer, 347 U.S. at 229-30, 74 S.Ct. 450; Mattox, 146 U.S. at 150-51, 13 S.Ct. 50.
[948]*948D.
Once a defendant shows an external occurrence having a tendency toward prejudice, federal law clearly requires a trial court to investigate the harmlessness or actual prejudice of the occurrence. Mattox, 146 U.S. at 150, 13 S.Ct. 50; Smith, 455 U.S. at 215, 102 S.Ct. 940 (“This Court has long held that the remedy for allegations of juror partiality is a hearing in which the defendant has the opportunity to prove actual bias.”). The Mattox Court categorically mandated that “possibly prejudicial” external contacts “invalidate the verdict, at least unless their harmlessness is made to appear.” Mattox, 146 U.S. at 150, 13 S.Ct. 50. This is required even if, as noted above, the contact did not constitute a communication nor concern a matter pending before the jury. See Smith, 455 U.S. at 215, 102 S.Ct. 940; Mattox, 146 U.S. at 150, 13 S.Ct. 50. Supreme Court case law also requires this, procedure irrespective of whether or not the court- knows “what actually transpired” and when, as the dissent highlights, the influence of that contact is speculative or uncertain.9 Remmer, 347 U.S. at 229, 74 S.Ct. 450; see also id. at 229-30, 74 S.Ct. 450 (mandating an eviden-tiary hearing where “information ... [about an,external contact] was received” by a trial court (emphasis added)). The Supreme Court has further held that prejudice is more probable where the record reflects that a jury could not agree as to the defendant’s guilt. Parker, 385 U.S. at 365, 87 S.Ct. 468 (citing as evidence of prejudice the fact that “the jurors deliberated for 26 hours, indicating a difference among them as to the guilt of petitioner”).
III.
In sum, the governing Supreme Court case law can be distilled as follows: Where a court receives information, Remmer, 347 U.S. at 229-30, 74 S.Ct., 450, about an unauthorized external contact between a juror and a government agent whose official position “carries great weight with a jury,” Parker, 385 U.S. at 365, 87 S.Ct. 468, that contact has á “tendency to ... influence” the verdict, and the trial court must presume the external contact prejudiced the defendant unless the government provides contrary evidence. Mattox, 146 U.S. at 150, 13 S.Ct. 50. This is true whether or not the contact was intentional, Gold, 352 U.S. at 985, 77 S.Ct. 378, whether or not the contact involved a verbal communication, Smith, 455 U.S. at 212, 102 S.Ct. 940; Mattox, 146 U.S. at 150, 13 S.Ct. 50, and whether or not the trial court or defendant “know[s] ... what actually transpired,” Remmer, 347 U.S. at 229, 74 S.Ct. 450.10 Once a potentially prej-[949]*949udieial contact is alleged; the court should “determine the circumstances, the impact thereof upon the juror, and whether or not it was prejudicial, in a hearing with -all interested parties permitted to participate.” Id. at 230, 74 S.Ct. 450.
In this case, the Nevada Supreme Court assumed a contact — albeit not a “communication” — occurred. Our case law compels our conclusion that the contact in question had enough of a tendency to influence the jury’s verdict so as to necessitate judicial inquiry into prejudice. It was thus, error for the Nevada Supreme Court not to conduct a prejudice analysis merely because Juror No. 2’s police tail did not amount to a “communication .,. about a matter pending before the jury.”
A,
Assuming the truth of Juror No. 2’s testimony that he had been followed closely for seven miles on the second day of deliberations,11 the Nevada Supreme Court concluded that this conduct does not constitute a “communication.” On this basis, the Nevada Supreme Court then concluded that any influence was “too speculative” to warrant examination of prejudice. Thus, the Nevada Supreme Court declined to consider whether the conduct in fact influenced the verdict. As set forth above, this decision contravenes the standard clearly established by Supreme Court case law, under which a defendant need not prove a “communication ... about a matter pending before the jury,” or even a “communication” about an unrelated issue. See Smith, 455 U.S. at 212-15, 102 S.Ct. 940. Only a threshold showing of any “contact,” Remmer, 347 U.S. at 229, 74 S.Ct. 450, with a “tendency to adverse influence” is required to' prompt the court to investigate whether that contact was, in fact, prejudicial. Mattox, 146 U.S. at 150, 13 S.Ct. 50.
In light of this, we have little trouble concluding that the contact that the Nevada Supreme Court assumed occurred had enough potential for ‘prejudice to cross Mattox’s low threshold. Las Vegas police officers were deeply entangled in this case as victims,’ witnesses, investigators, and trial spectator's. Juror No, 2 testified that he had been closely followed’ by a marked police car for over sevén miles. See Parker, 385 U.S. at 365, 87 S.Ct. 468 (observing that government agents “carr[y] great weight with a jury”). Juror No. 2’s testimony indicates that the tail was maintained at a distance so close that Juror No. 2 could not see the police vehicle’s front wheels or bumper — if true, this conduct could have reasonably been understood as an attempt to intimidate. Moreover, Juror No. 2 was a known hold-out before the contact occurred. The Supreme Court has clearly established that the likelihood of possible prejudice increases where, as herd, the jury was previously deadlocked. See id.
Thus, because the state court assumed that the contact did in fact occur and clear[950]*950ly established case law demonstrates that the contact had a tendency to affect the verdict, the court should have, at a minimum, investigated the prejudice or harmlessness of the contact even if at the time the court was unaware what exactly transpired or whether the impact was harmful. See Remmer, 347, U.S. at 229, 74 S.Ct. 450. The Nevada Supreme Court erred when it failed to do so.
Certainly, there may be circumstances in which a trial court finds a juror’s allegations of an external contact are unsupported by sufficient evidence, or in which the allegations are so implausible or incredible that they may be reasonably disregarded. There may also be cases in which an alleged external contact suggests paranoia or some underlying mental incompetence on the juror’s part. See Tanner, 483 U.S. at 118-19, 107 S.Ct. 2739. Under those circumstances, a court will not run afoul of the Constitution by refusing to consider whether the alleged contact affected the verdict. But this is not the case on the record before us.
Here, Juror- No. 2’s testimony was not discredited. To the contrary, crediting Juror No. 2’s testimony about a plausible external contact with a juror reluctant to convict, the Nevada Supreme Court declined to consider whether Juror No. 2 may have been prejudiced by the police tail.12 This contravened clearly established federal law.13 See Remmer, 347 U.S. at 229-30, 74 S.Ct. 450; Mattox, 146 U.S. at 150, 13 S.Ct. 50; Smith, 455 U.S. at 215, 102 S.Ct. 940.
IV.
Because the Nevada Supreme Court failed to consider the prejudicial impact of the contact, in violation of the law clearly established in Mattox, we may evaluate Tarango’s claim “without deference to the state court’s decision” and “unencumbered by the deference AEDPA normally requires.”14 Panetti, 551 U.S. at 948, 127 S.Ct. 2842; see also Castellanos, 766 F.3d at 1146. Reviewing de novo, we hold that the Nevada trial court improperly restricted the scope of the evidentiary hearing, effectively preventing Tarango from proving prejudice.
Under our precedent, where an external contact with the jury is shown, a trial court should determine whether the contact “raises a risk of influencing the verdict.” Caliendo, 365 F.3d at 697. Under [951]*951such circumstances, prejudice is presumed and the government bears the burden of rebutting the presumption of prejudice. Id. To be sure, “certain chance contacts between witnesses and jury members — while passing in the hall or crowded together in an elevator — may be inevitable.” Id. at 696 (internal quotation marks and citation omitted). Therefore, if the contact involves a “prosaic” or “more common and less pernicious extraneous influence” than jury tampering, the court should determine whether the jury was “substantially swayed” by the contact.15 United States v. Henley, 238 F.3d 1111, 1115-16 (9th Cir, 2001). Under this circumstance, the defendant bears the burden of offering sufficient evidence to trigger a presumption of prejudice. See Caliendo, 365 F.3d at 696-97 (collecting authorities).
Although the anti-impeachment rule, codified as Federal Rule of Evidence 606(b)(1), prohibits juror testimony regarding “any juror’s mental processes concerning the verdict,” an exception to the rule permits juror testimony about whether “an outside influence was improperly brought to bear on any juror.” Fed. R. Evid. 606(b)(2)(B). This court has accordingly deemed admissible limited juror testimony to determine “the impact [of an outside influence] upon the juror, and whether or not [the outside influence] was prejudicial.” Remmer, 347 U.S. at 230, 74 S.Ct. 460; see also Rutherford, 371 F.3d at 643-45 (considering juror affidavits including claims that the jury felt intimidated by police officers’ glares); Caliendo, 365 F.3d at 699 (considering a juror’s testimony that the jury’s external communication with a police officer left them with a favorable opinion of the officer).
Unlike Nevada law, our precedent instructs, that a court should not limit juror testimony to “the existence of [an external contact].” Rutherford, 371 F.3d at 644 (quoting, the district court in that case). Rather, a court “should [also] consider the ‘effect of extraneous information or improper contacts on a juror’s state of mind,’ a juror’s ‘general fear and anxiety following’ such an incident, and any other thoughts a juror might have about the contacts or conduct at issue.” Id. (quoting United States v. Elias, 269 F.3d 1003, 1020 (9th Cir. 2001)). To that end, consistent with the anti-impeachment rule, this court permits the introduction of limited evidence of a juror’s state of mind to prove juror misconduct. A court may not, consistent with the anti-impeachment rule, admit testimony “regarding the affected juror’s mental processes in reaching the verdict.” Id. (internal quotation; marks omitted) (quoting Elias, 269 F.3d at 1020). However, “a juror’s testimony concerning his fear that individuals would retaliate against him if he voted to acquit (or convict) would be admissible, although his statement that he actually cast his vote one way or the other because of that fear would not.” Id.
Consistent with the principles announced in Rutherford, the district court should admit Juror No. 2’s. statements about how the police tail impacted him, although not how it impacted his deliberations and verdict. Therefore, Juror No. 2’s statement that he found the police tail “unnerving” is admissible, as are his statements that he “concluded Metro somehow knew who [he] was.” By contrast, Juror No. 2’s statements that he “relinquished his vote under duress,” and “still [has] doubt as an X-Juror” are not admissible.
[952]*952V.
Because the scope of the evidentiary-hearing was narrowly circumscribed in the state trial court, the record before us is insufficient to determine whether the police tail influenced the verdict and preju-dicéd Tarango. We accordingly remand for the district court to hold an evidentiary hearing and apply the proper standard to determine whether the Nevada courts violated Tarango’s due process right to a fair and impartial jury by failing to adequately consider allegations of a prejudicial external influence on the jury. Following this court’s precedent, the district court should permit Tarango to offer limited evidence to show prejudice, see Caliendo, 365 F.3d at 696-97; Henley, 238 F.3d at 1115-16, including evidence of Juror No. 2’s “general fear and anxiety” following the police tail, see Rutherford, 371 F.3d at 644. VACATED AND REMANDED.